RAVE is a variational autoencoder designed for fast, high-quality neural audio synthesis. It allows training on custom audio datasets to generate new sounds in real-time. The model can be used as a VST plugin for integration with music production software, and supports both model training and inference workflows.